Chris Collins teaches HR analytics at Cornell’s School of Industrial and Labor Relations. He lets his students use AI to run the numbers. That is a deliberate choice, and it says something important.
“The real value is not in running the analysis,” Collins told me. “It is in understanding what the data means and how it should inform decisions. That is where human judgment remains essential.”
Collins is a leading scholar in strategic HR and talent systems, and when I interviewed him about generative AI’s role in the function, he was precise about something most of the conversation is not: AI’s value in HR is real and already here, and it exists almost entirely in data aggregation, not in the work that actually defines the function.
That distinction matters more than most organizations currently appreciate.
See also: The AI revolution is here: How to optimize AI for HR
Where AI is already working
The performance gap AI closes most immediately is speed and comprehensiveness. Compensation benchmarking across regions, pay equity analysis, workforce trend identification, routine employee inquiries—all of these are areas where AI can do in minutes what used to take people days.
“Running numbers, benchmarking compensation and aggregating regional pay data are all things AI can do extremely well,” Collins says. “The speed and access to information it provides can be very effective for HR leaders.”
This is not small. HR decisions have long been constrained by slow or incomplete data. Generative AI removes many of those barriers, and the efficiency gains are substantial. But Collins draws a clean line between informing judgment and replacing it.
The bias problem nobody is solving
The most immediate risk Collins flagged is one that receives attention in the abstract and insufficient scrutiny in practice. AI systems trained on historical HR data inherit historical HR decisions, which carry embedded bias.
“AI is drawing on prior decisions and prior performance data,” Collins says. “That means it can pull existing biases forward and, in some cases, actually amplify them. AI should be augmenting decisions, not making them. Leaving important people decisions entirely to AI is problematic.”
What makes this harder is that the bias is often invisible at the system level. It shows up in who gets flagged, who gets passed over and who receives a development opportunity, and the pattern only becomes visible in aggregate, often after significant harm has already been done.
The capability loss no one is talking about
This was the sharpest part of Collins’ analysis, and the one most absent from public discussion.
As AI absorbs routine and repetitive tasks, some organizations are discovering they no longer have strong capabilities in areas that were once foundational to HR: job design, workflow analysis and organizational design. Those skills atrophied because they were not in demand.
Now, as organizations face pressure to redesign work around AI, those skills are exactly what is needed.
“We are seeing organizations talk about people doing different jobs, but HR often lacks strong job and workflow design capabilities,” Collins says. “When those skills are missing, the results can be uneven and lumpy across the organization.”
He raised a related concern about implementation. Many organizations are layering AI tools onto existing HR processes without rethinking the overall system. The result is fragmented, and employees feel it.
“What worries me is the unevenness of the experience employees are getting. When AI is added without rethinking how the system works end to end, it can create disjointed and frustrating experiences rather than improving them.”
The roles that will disappear
Collins was direct about this: Some HR roles will not survive the shift, not because AI is threatening, but because data will increasingly flow directly to line leaders. When AI surfaces pay gaps, performance trends or workforce risks in real time, managers no longer need HR to retrieve that information.
That changes what HR is for. The function’s value proposition shifts from reporting to interpretation and strategic guidance. That is a different kind of HR professional, and not every organization is developing one.
The same logic applies to compensation conversations. AI can model scenarios and surface inequities faster and more comprehensively than traditional approaches. But Collins was clear that data alone does not produce better outcomes.
“Simply giving someone a number does not lead to better outcomes. The discussion that goes with compensation decisions still matters. AI can support that conversation, but it cannot replace it.”
What AI cannot do
Coaching is where Collins is most skeptical of the current trajectory. Effective coaching is not information retrieval. It depends on helping someone make sense of their own situation, which requires empathy and context that AI lacks.
“A lot of coaching is about helping someone make sense of their own situation and arrive at their own solution. AI is good at sorting information, but it struggles with emotion, sentiment and context. Those are very human capabilities.”
The same constraint applies more broadly. Collins made an observation that is easy to overlook: “AI performs well in stable, repeating environments. When conditions change, it has to be retrained. Human systems adapt continuously.”
Employee experience reflects this. It is shaped by leadership, relationships, purpose and daily interaction. No AI layer changes the underlying system.
What this means for leaders
The organizations most at risk are not the ones that refuse to adopt AI. They are the ones who adopt it without design. Adding AI to a broken process produces a faster, more efficient broken process. Adding AI without maintaining human interpretive capability produces outputs that no one can act on well.
The function AI is most likely to transform is not the tactical work it is replacing. It is the strategic work that is raising expectations for it. As AI handles more of the routine, leaders will have less tolerance for HR that cannot interpret, design and guide. That bar is going up.
Collins puts it plainly: “AI is very good at repeated tasks in stable environments. But when things change, it has to be retrained. Human systems adapt continuously.”
The future of HR is not a question of people versus technology. It is a question of whether HR leaders understand what each does well, and whether they have built the organizational capability to use both.
Those who have not will find that AI has not replaced them. It just made the gap more visible.
Credit: Source link
